The international diplomatic corps has found its new favorite playground, and it goes by the name of AI governance.
Watch any United Nations panel or international summit, and you will see a familiar theater. High-ranking diplomats, including Canada's prominent envoys, take the stage to declare that managing artificial intelligence is the defining multilateral challenge of our generation. They call for global treaties, centralized oversight bodies, and harmonized regulatory frameworks. They treat AI as if it were a rogue nuclear arsenal waiting for a non-proliferation treaty.
It is a beautiful, expensive, and utterly useless illusion.
The current obsession with top-down, global AI regulation is not just slow—it is fundamentally incompatible with how software is built, deployed, and weaponized. While well-meaning bureaucrats spend years debating definitions of algorithmic bias in Geneva, open-source models are cloned millions of times on GitHub in seconds. The lazy consensus states that international diplomacy can corral compute power. The reality is that trying to govern AI through the UN is like trying to regulate the printing press by asking 193 countries to agree on what people are allowed to write.
The Flawed Analogy of the Nuclear Model
The biggest mistake global regulators make is treating AI like nuclear technology.
Nuclear governance worked—barely—because the underlying physics imposed massive material constraints. To build a weapon, you need enriched uranium or plutonium. You need centrifuges, massive cooling towers, and highly visible industrial infrastructure. A state cannot hide a nuclear program in a basement or on a thumb drive. International bodies like the International Atomic Energy Agency (IAEA) can monitor physical supply chains because those chains are bottlenecked by rare earth elements and massive industrial footprints.
AI laughs at these constraints.
[Traditional Geopolitics: Nuclear]
Enriched Material ➔ Physical Facilities ➔ State Control ➔ Verifiable Treaties
[Modern Reality: Advanced AI]
Commodity Chips ➔ Decentralized Clusters ➔ Open-Source Weights ➔ Impossible Verification
To build a world-class foundational model, you do need heavy capital: tens of thousands of specialized graphical processing units (GPUs) clustered together. But once that model is trained, the game changes entirely. The resulting file—the model weights—can be compressed, leaked, and run on consumer-grade hardware or decentralized networks.
When Meta leaked its LLaMA weights in 2023, it did not take a state-sponsored laboratory to adapt them. Within days, independent developers optimized the code to run on a MacBook.
How does an international treaty monitor a file sitting on a developer's local hard drive in Buenos Aires or Bucharest? It cannot. Any governance framework built on the assumption that you can audit, track, or lock down code at a global scale is dead on arrival.
Bureaucracy Moves in Years. Code Moves in Seconds.
I have spent years watching enterprise organizations and state actors attempt to map traditional compliance models onto software development. I have seen institutions burn millions of dollars hiring consultancy firms to build massive framework documents that are obsolete before the ink dries.
The mismatch is a matter of basic physics and operational velocity.
Diplomacy is intentionally slow. It relies on consensus, semantic debates, and ratification processes that span years. Consider the European Union's AI Act. By the time it crawled through the legislative meat grinder and reached finalization, the entire technological landscape had shifted from basic predictive algorithms to generative multimodal systems that can code, reason, and generate video. The law was written for yesterday’s calculators and applied to today’s engines.
When states attempt to center their AI strategy around global forums like the UN, they are effectively outsourcing their national security to a committee.
The hard truth nobody in diplomacy wants to admit is that international bodies have zero enforcement teeth against the actors who matter most. If the United States, China, and a handful of private tech monopolies decide how they want to deploy their models, a UN resolution will not stop them. It merely binds the hands of middle powers who choose to follow the rules, handicapping their own domestic tech sectors while adversaries operate with total impunity.
Dismantling the "People Also Ask" Myth of Safe AI
If you look at public forums or regulatory white papers, the same questions appear on loop. The premises of these questions are fundamentally broken.
Can we create a global framework to eliminate algorithmic bias?
No. And stop trying. The idea that a single global standard can define "fairness" is a philosophical absurdity. A content moderation or hiring algorithm optimized for the cultural norms of San Francisco will violate the laws or cultural values of Singapore, Riyadh, or Warsaw.
When international bodies try to harmonize these standards, they end up with bland, unworkable guidelines that mean everything and nothing. Fairness is not a technical problem to be solved with a global patch; it is a political compromise that varies by jurisdiction.
Won't unregulated AI lead to existential risk?
The real existential risk is not a sentient superintelligence turning humanity into paperclips. The risk is structural fragility.
By obsessing over sci-fi scenarios, global governance bodies ignore the immediate, mundane threats: automated financial trading systems creating flash crashes, deepfake supply chains eroding institutional trust, and autonomous cyber-weapons targeting infrastructure. These threats do not care about high-level ethical declarations. They require hard-nosed technical resilience, cyber defense, and redundant physical infrastructure—not another non-binding declaration signed in New York.
The Cost of the Contrarian Reality
If we accept that global governance is an illusion, what is the alternative? The alternative is messy, fragmented, and highly competitive. It is an approach called Localized Containment and Compute Asymmetry.
Instead of chasing the fantasy of global consensus, nations must focus on what they can actually control:
- Physical Compute Chokepoints: The only piece of the AI stack that behaves like a nuclear asset is the semiconductor manufacturing process. ASML's extreme ultraviolet (EUV) lithography machines, built in the Netherlands, are the true levers of control.
- Domestic Liability Frameworks: Stop trying to regulate the technology itself. Regulate the outcomes. If an autonomous vehicle crashes, or a medical diagnostic tool misdiagnoses a patient, the existing legal framework of tort law, product liability, and corporate negligence should apply. We do not need an "AI Law" for this; we need to hold the deployers of software accountable for the damages they cause.
- Redundant Defense: Assume that bad actors will get their hands on powerful, unrestricted models. The strategy cannot be prevention; it must be mitigation. This means building defensive AI systems designed to catch deepfakes, secure critical infrastructure, and monitor financial anomalies in real-time.
| Strategic Approach | Core Philosophy | Primary Mechanism | Enforcement Capability |
| :--- | :--- | :--- | :--- |
| *Global Governance* | Universal Consensus | Treaties & UN Panels | Non-binding, unenforceable |
| *Compute Asymmetry* | Supply Chain Control | Semiconductor Export Caps | High (Physical monopolies) |
| *Localized Liability*| Outcome Accountability | Existing Courts & Tort Law | High (Domestic enforcement) |
The downside to this realistic view is obvious: it guarantees an AI arms race. It means the dream of a harmonious, unified digital world is dead. It means accept fragmented regional standards, where the internet splits further into Western, Chinese, and non-aligned blocs.
But hiding from this reality behind diplomatic pleasantries is actively dangerous.
Nations that spend the next five years focused on drafting international statements will find themselves defenseless against adversaries who spent that same time scaling their compute infrastructure and integrating models into their national defense apparatus.
Stop Committing Diplomacy to a Software Problem
The urge to govern AI from the heights of international diplomacy is an admission of fear. It is the reaction of traditional institutions realizing they no longer control the speed of innovation, attempting to slow the world down to the speed of a committee meeting.
Canada and its allies should stop treating the UN as the primary arena for AI strategy. It is a secondary theater for optics, not outcomes.
If a nation wants a seat at the table in the era of artificial intelligence, it does not earn it by hosting summits or chairing subcommittees on ethics. It earns it by building sovereign compute capacity, securing its critical technology supply chains, and training engineers who understand the code deeply enough to defend against it.
The future is not going to be regulated by a treaty. It is going to be written in code, run on silicon, and secured by those who stop talking and start building.
